Skip to main content

Getting Started

Prerequisites​

  • Bun ≥ 1.0
  • A TypeScript or JavaScript project with a git repository

Installation​

# Clone the repo
git clone https://github.com/wellbrito29/braito.git
cd braito

# Install dependencies
bun install

First run​

bun run scan        # list all eligible files
bun run generate    # generate notes (static-only, no LLM)

This creates a .ai-notes/ directory with one .json + .md sidecar per file, plus an index.json and index.md summary.

Enable LLM synthesis​

Set your API key and configure a provider in ai-notes.config.ts:

// ai-notes.config.ts
export default {
  llm: {
    provider: 'anthropic',          // 'anthropic' | 'openai' | 'ollama' | 'claude-cli'
    model: 'claude-sonnet-4-6',
    llmThreshold: 0.4,              // only files above this score get LLM synthesis
  },
}
export ANTHROPIC_API_KEY=sk-ant-...
bun run generate

Files with criticalityScore >= llmThreshold are sent to the LLM. All others get a fast static note.

CLI commands​

ScriptEquivalent flagsDescription
bun run scan--format jsonDiscover and list eligible files
bun run generate--filter --diff --languageFull pipeline — writes .ai-notes/
bun run generate:force—Bypass cache, reprocess all files
bun run generate:dry—Preview without writing files
bun run generate:v—Per-file signal detail + phase timers
bun run watch—Regenerate on file change
bun run mcp--auto-generateMCP server (JSON-RPC 2.0 over stdio)
bun run ui--port <n>Local web UI at http://localhost:7842
bun run init:agent—Generate .claude/commands/ slash command files
bun test—Run all test suites

Generated output​

.ai-notes/
  src/
    core/
      scanner/discoverFiles.ts.json   ← structured note
      scanner/discoverFiles.ts.md     ← human-readable sidecar
  index.json                          ← all files ranked by criticality
  index.md                            ← grouped by domain

cache/
  hashes.json                         ← SHA-1 per file for incremental runs

Do not edit .ai-notes/ or cache/ manually — they are regenerated on each run.

Next steps​